Digital twin-driven drying system and energy flow-quality cooperative control method
By using a digital twin-driven CO2 heat pump drying system, combined with multi-source data sensing and the PINN-Transformer model, energy flow and quality collaborative control is achieved. This solves the problems of energy flow mismatch and quality fluctuation in traditional CO2 heat pump drying systems under complex operating conditions, enabling efficient heating, precise humidity control, and full-process traceability, thus meeting the intelligent management and control requirements of modern production.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- YUNNAN NORMAL UNIV
- Filing Date
- 2026-03-23
- Publication Date
- 2026-05-12
AI Technical Summary
Existing CO2 heat pump drying systems suffer from problems such as energy flow mismatch, insufficient coordination between dehumidification and heating, lagging regulation, and product quality fluctuations under conditions of severe cold and low temperature, large temperature rise, and alternating temperature and humidity. They lack the ability to model and accurately perceive the dynamic evolution of temperature and humidity fields in multiple regions, material dryness gradients, and quality, and cannot achieve on-demand energy supply and precise humidity control. Furthermore, they lack full-link closed-loop control driven by digital twins.
The drying system driven by digital twins combines multi-source heterogeneous data perception, PINN-Transformer coupled digital twin model, machine vision and reinforcement learning to achieve coordinated control of energy flow and quality. Through multi-physics coupling modeling and dynamic attention mechanism, real-time data fusion and decision optimization are carried out, and cloud-edge collaborative computing and trusted evidence storage mechanism are constructed to achieve efficient heating, precise humidity control and consistent drying quality.
It enables optimized energy flow allocation and precise quality control under complex operating conditions, improves drying uniformity and product quality consistency, adapts to the response accuracy of extreme environments, meets the intelligent control requirements of modern production, and realizes a digital closed loop for full-process traceability and management.
Smart Images
Figure CN122015462A_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the fields of heat pump drying, industrial internet and intelligent manufacturing technology, and in particular relates to a digital twin-driven drying system and a method for coordinated control of energy flow and quality. Background Technology
[0002] Carbon dioxide heat pump drying has attracted attention due to its environmentally friendly working fluid and low-temperature adaptability. However, in severe cold and low temperature heating conditions, as well as in environments with alternating or fluctuating temperature and humidity, the state of the drying medium and the moisture content field of the material exhibit strong nonlinear changes, which can easily lead to problems such as energy flow mismatch, insufficient coordination between dehumidification and heating, lag in regulation, and fluctuations in product quality.
[0003] Existing drying equipment control strategies largely rely on fixed threshold regulation or single-loop feedback control modes, lacking real-time online modeling and precise sensing capabilities for multi-regional temperature and humidity fields, material dryness gradients, and the dynamic evolution of quality. Even though some technologies attempt to optimize control logic and use PLCs to achieve quality feedback control for CO2 heat pump drying, they remain limited to a single closed-loop cycle and a single quality target, lacking coordinated optimization design for energy flow and quality. While some technologies mention balancing energy efficiency and quality, they only design discrete segmented control for rice grains, lacking both precise energy flow allocation models and the ability to adapt to the characteristics of heat-sensitive materials. This makes it difficult for existing technologies to achieve dynamic matching of on-demand energy supply and precise humidity control when facing complex conditions such as environmental disturbances, load fluctuations, and material anisotropy, further exacerbating the management contradiction between energy efficiency improvement and quality assurance.
[0004] Meanwhile, with the deep penetration of intelligent manufacturing technology, the demand for standardized management of formulation processes, full-process batch traceability, and remote operation and maintenance control in drying production lines is becoming increasingly urgent, but existing technologies still have significant shortcomings. At the digital twin application level, it is limited to unidirectional optimization in the R&D stage, lacking real-time control capabilities for the production process; it also lacks deep modeling and decision-making functions for energy flow and quality parameters. At the heat pump adaptation and collaborative control level, it focuses on air-source heat pump temperature and humidity control, without addressing CO2 transcritical cycle conditions, and lacks closed-loop design for quality parameters, failing to meet the multi-objective control requirements under complex operating conditions. In summary, traditional CO2 heat pump drying systems generally lack multi-source data fusion perception, digital twin reasoning and decision-making, and reliable data storage mechanisms, failing to establish a complete closed-loop system encompassing "energy flow efficiency - product quality - process traceability," and thus struggling to adapt to the intelligent control requirements of modern production.
[0005] Based on the multiple bottlenecks of the existing technologies, there is an urgent need to propose a digital twin-driven intelligent drying system and control method for carbon dioxide heat pumps, which is based on data-driven and multi-physics field coupled modeling, can adapt to temperature and humidity alternating disturbances, and achieve coordinated energy flow optimization and precise quality control. Summary of the Invention
[0006] The purpose of this invention is to provide a digital twin-driven drying system and an energy flow-quality collaborative control method, which solves the problems in existing technologies such as the general lack of multi-source data fusion sensing, digital twin reasoning decision-making, and reliable data storage mechanisms in traditional CO2 heat pump drying systems, the inability to establish a complete closed loop of "energy flow efficiency-product quality-process traceability", and the difficulty in adapting to the intelligent management and control requirements of modern production. This invention achieves efficient heating, precise humidity control, and consistent drying quality under complex operating conditions through the synergy of physical systems, multi-source sensing, digital twins, intelligent optimization control, and reliable traceability.
[0007] To achieve the above objectives, the present invention provides a digital twin-driven drying system, comprising: a carbon dioxide heat pump power supply subsystem, an air circulation and heat recovery subsystem, a drying device, and an intelligent control and digital management subsystem; The carbon dioxide heat pump energy supply subsystem includes a compressor, an air cooler, a throttling element, and an evaporator, which is used to provide heat to the drying unit and form a heat exchange coupling with the air circulation and heat recovery subsystem; The air circulation and heat recovery subsystem includes a return air duct, a fresh air duct, a dehumidification duct, a waste heat recovery unit, a blower, a return air blower, a dehumidification blower, and a stratified air supply actuator, which is used to recover waste heat and condense and dehumidify the return air and supply air to the drying unit in stratified manner. The intelligent control and digital management subsystem is electrically connected to the carbon dioxide heat pump energy supply subsystem and the air circulation and heat recovery subsystem to generate coordinated control commands.
[0008] Preferably, the intelligent control and digital management subsystem includes a multi-source heterogeneous process data perception and fusion module, a machine vision and moisture content prediction module, a PINN-Transformer coupled digital twin model module, an energy flow-quality collaborative intelligent regulation module, a disturbance adaptive operating condition identification and mode switching module, a dynamic attention mechanism module, a cloud-edge collaborative computing module, and a reliable operation data storage module. The output of the PINN-Transformer coupled digital twin model module includes air state parameters, spatial distribution of material moisture content, energy consumption per unit of moisture removal, drying rate, and quality risk indicators, which are used to provide feedforward predictions for the energy flow-quality collaborative intelligent control module. The dynamic attention mechanism module is used to jointly model the multi-layer / multi-zone temperature and humidity field, air volume distribution, and material moisture content distribution to form spatiotemporal features for digital twin prediction and control optimization. The cloud-edge collaborative computing module includes an edge controller deployed at the drying site and a cloud server; the edge controller is used to perform data fusion, real-time inference of digital twins, and control output; the cloud server is used to perform model training, recipe optimization, and global parameter tuning, and synchronizes parameters with the edge controller. The trusted evidence storage module for operational data is used to generate hash values for operational data, quality data, and control decisions, and write them into the consortium blockchain or private blockchain ledger in chronological order for verification and traceability. The intelligent control and digital management subsystem interfaces with SCADA systems, MES systems, or industrial IoT platforms via OPC UA, Modbus, or MQTT protocols for process recipe distribution, batch management, and equipment status monitoring.
[0009] Preferably, the multi-source heterogeneous process data sensing and fusion module includes a temperature sensor, a relative humidity sensor, a wind speed / air volume sensor, a pressure sensor, an online moisture content sensor, an energy consumption electrical parameter acquisition unit, and a material weight acquisition unit, used to acquire the operating status and material status data of the carbon dioxide heat pump power supply subsystem and the air circulation and heat recovery subsystem.
[0010] Preferably, the machine vision and moisture content prediction module includes an industrial camera and an edge inference unit, which uses a target detection and texture feature extraction model to identify the appearance features of the material and outputs online prediction results of the material's moisture content and drying uniformity.
[0011] Preferably, the energy flow-quality collaborative intelligent control module generates a hierarchical energy flow priority allocation scheme based on a reinforcement learning strategy, and combines model predictive control to constrain and optimize the compressor frequency, throttling element opening, fan speed, hierarchical air supply valves, and delivery cycle.
[0012] Preferably, the disturbance adaptive operating condition identification and mode switching module constructs disturbance indices based on environmental temperature and humidity fluctuations, material moisture content gradients, and energy flow distribution offsets, and adaptively switches between closed-loop circulation, open-loop dehumidification, deep condensation dehumidification and energy replenishment, and frost suppression / defrosting scheduling modes, while imposing constraints on pressure change rate, exhaust temperature, and heat supply fluctuations.
[0013] A method for coordinated energy flow and quality control, applying the aforementioned digital twin-driven drying system, includes the following steps: S1. Collect operating parameters of the carbon dioxide heat pump power supply subsystem, temperature, humidity and air volume parameters of multiple layers / zones in the drying device, energy consumption parameters and material status data, and perform data cleaning and synchronization. S2. Perform target detection and texture feature extraction on the collected image data to obtain the material appearance features and output the online prediction results of material moisture content and drying uniformity; S3. Use dynamic attention mechanism to perform spatiotemporal feature fusion on the multi-source data obtained in S1 and S2 to form a feature vector for twin reasoning; S4. Based on the constraints of heat and mass transfer and energy conservation, drive the PINN-Transformer coupled digital twin model to predict air state parameters, spatial distribution of material moisture content, energy consumption per unit of moisture removal, and quality risk. S5. Based on the predicted values obtained in S4, construct a multi-objective optimization problem of energy consumption, humidity control, and quality; screen key control variables through feature importance analysis, and establish a mapping relationship between heat pump COP, material apparent characteristics, and macroscopic drying rate, and dynamically adjust the weights of the multi-objectives. S6. Based on the cooperative strategy of reinforcement learning and model predictive control, hierarchical energy flow priority allocation and multi-actuator cooperative control commands are generated, and operating conditions are identified and modes are switched according to disturbance indicators. S7. Reliably store operational data, quality data, and control decisions, and associate them with batch and formulation information.
[0014] Preferably, the specific content of the PINN-Transformer coupled digital twin model driven by heat transfer, mass transfer and energy conservation constraints in S4 is as follows: heat transfer, mass transfer and air state constraints are embedded in the data-driven network, and the material moisture content diffusion equation, surface convection mass transfer boundary conditions, zoned air energy conservation relationship and air enthalpy-humidity relationship are used as physical constraint equations in the training and updating process of the digital twin model, so as to achieve high-precision coupled prediction of temperature field, humidity field, moisture content evolution and system energy flow state during the drying process; The diffusion equation for the moisture content of the material is expressed as follows: ; In the formula, Moisture content of the material; The equivalent diffusion coefficient; This is the spatial second derivative of the moisture content, representing the diffusion and transfer of moisture within the material. The surface convection mass transfer boundary condition is expressed as follows: ; In the formula, The moisture content gradient is along the direction of the outer normal to the material surface. The surface convective mass transfer coefficient; Moisture content on the surface of the material; To achieve equilibrium moisture content; No. The energy of air in the dry zone is conserved, as expressed below: ; In the formula, For the firstj Air mass flow rate in the dry zone; The specific heat capacity of dry air at constant pressure; and The first j Inlet and outlet air temperatures of the drying zone; For the first j Heat exchange in the drying zone; The latent heat of vaporization of water; For the first j Mass flow rate of water evaporated in the drying zone; The enthalpy-humidity relationship of air is expressed as follows: ; ; In the formula, Moisture content of humid air; It is the partial pressure of water vapor; This is the total pressure of the moist air; Specific enthalpy of moist air; The specific heat capacity of dry air at constant pressure; Air temperature; The latent heat of vaporization of water at a reference temperature of 0℃; The specific heat capacity of water vapor at constant pressure; The prediction model is obtained by jointly optimizing the data fitting loss and the physical residual loss. The loss function is constructed as follows: ; In the formula, This is the total loss function; This is the loss term for data fitting; This is the physical residual loss term; This is the regularization loss term; This is the weighting coefficient for physical residual loss; The regularization loss weight coefficient; Among them, data fitting loss The expression is as follows: ; In the formula, The number of samples; For the first i Predicted values of each output quantity; For the first i Measured values of each output quantity; Among them, physical residual loss The expression is as follows: ; In the formula, Number of physical configuration points; For physically constrained residual operators; For the first j Predicted physical field variables for each configuration point; For the first j The spatial coordinates of each configuration point; For the first j The time coordinates of each configuration point; Among them, the smoothing regularization term The expression is as follows: ; In the formula, For the model k One trainable parameter; This represents the total number of trainable parameters. Set the discrete control period Δ t The digital twin model uses a state-space representation for online assimilation, expressed as follows: ; ; In the formula, for t The system state vector at any given time; for t Control the quantity at all times; for t The amount of disturbance at any given moment; for t The system output vector at each time step; This is the state transition function; For observation mapping function; The PINN-Transformer coupled digital twin model employs an adaptive weighting strategy for physical residuals to adjust the weights of different physical process constraints online, maintaining prediction stability under conditions such as condensation, strong dehumidification, or sudden load changes.
[0015] Preferably, the specific details of dynamically adjusting the multi-objective weights in S5 are as follows: The energy flow-quality collaborative intelligent control module integrates heat pump thermodynamic parameters, drying oven temperature and humidity field, and material dryness gradient to construct a deep learning-driven energy flow control model; it designs a hierarchical energy flow priority allocation strategy based on a reinforcement learning framework and performs multi-objective optimization. The expression for constructing the partition energy flow priority is as follows: ; In the formula, For the first j Energy flow priority index for the drying zone; For the first j Moisture content deviation or dryness gradient in the drying zone; For the firstj Quality risk indicators for the drying area; The heating demand index for the j-th dry zone; Weighting for moisture content deviation; As a quality risk weight; Weighting based on heating demand; j This is the number for the drying zone.
[0016] The objective function expression for model predictive control optimization is as follows: ; In the formula, The model predicts the control objective function; N To predict the length of the time domain; k For prediction steps; For the first t+k Step-by-step prediction output; For reference trajectory; Q The output error weighting matrix; For the first t+k Step-by-step control increment; R To control the incremental weighting matrix, This is the penalty coefficient for the variance of moisture content; For the first t+k The variance term for the predicted moisture content state parameters; For the first t+k Predicting moisture content state parameters; The constraints that the model predictive control optimization must satisfy are as follows: And meet the safety interlock constraints; In the formula, and These are the lower and upper limits of the control quantity, respectively; and These are the lower and upper limits for controlling the increment, respectively; The expression for constructing the reinforcement learning reward function is as follows: In the formula, Let t be the reward function value; Let t be the system energy consumption or the energy consumption per unit of water removal at time t. Energy consumption penalty weight; Let t be the average moisture content at time t; The target moisture content; This is the variance term for the moisture content of each drying zone; The quality penalty term at time t; Weighted for average moisture content tracking; The weight for the moisture content variance penalty; Weighting for quality penalties.
[0017] Preferably, the specific details of operating condition identification and mode switching based on disturbance indicators in S6 are as follows: A disturbance index is constructed based on temperature and humidity fluctuations and material moisture content gradients. When the disturbance index exceeds a preset threshold, the humidity control weight and quality weight are adaptively increased, and energy flow is preferentially allocated to areas with high moisture content to improve drying uniformity and quality consistency. The mode switching includes closed-loop mode, open-loop dehumidification mode, deep condensation dehumidification and energy replenishment mode, and defrost / defrost scheduling mode. During the mode switching process, constraints are imposed on the pressure change rate, exhaust temperature and heat supply fluctuations. The expressions for constructing the disturbance index and the frost risk index are as follows: ; In the formula, Let be the disturbance index at time t; The frost risk index at time t; The inlet air temperature; The relative humidity of the inlet air; For the measurement of air supply volume or wind speed; This refers to the surface temperature of the evaporator or heat exchanger. This corresponds to the dew point temperature; For indicator functions; , , , These are the weighting coefficients for each component of the disturbance index; The cumulative weighting coefficient for frost risk; , and These are the normalized reference values for temperature, relative humidity, and air volume, respectively. The sampling interval; For the time window of frosting accumulation; The threshold for the disturbance index; The threshold for frost risk; This is the duration threshold; When satisfied > And the duration exceeds When the conditions are met, the system switches to high-disturbance control mode; when the conditions are met... > When this happens, the system switches to either defrost suppression or defrost scheduling mode.
[0018] Therefore, the beneficial effects of the above-described system and method in this invention are as follows: (1) Based on multi-source heterogeneous data fusion and dynamic attention modeling, the spatiotemporal coupling relationship between temperature and humidity field, material dryness gradient and energy flow distribution is accurately characterized, effectively alleviating the prediction deviation and control lag caused by temperature and humidity alternation disturbance, overcoming the defect of single-loop control in insufficient adaptability to complex working conditions, and improving the system's response accuracy to extreme scenarios such as severe cold and low temperature, temperature and humidity fluctuation. (2) By constructing a digital twin coupled with PINN-Transformer, physical mechanism constraints such as heat transfer, mass transfer and energy conservation are embedded into the data-driven model, which solves the shortcomings of existing digital twin technology that only stays at simulation or one-way perception and lacks mechanism support, significantly improves the physical consistency of prediction results, and enhances the system's generalization ability and operational stability in extreme environments. (3) The integration of machine vision and online moisture content prediction technology realizes the visualization of drying status. Combined with reinforcement learning and model predictive control algorithm, an energy flow-quality collaborative optimization mechanism is constructed to realize the dynamic regulation of priority energy supply in high moisture content area and precise humidity control in low dryness area, which greatly improves the drying uniformity and product quality consistency, especially suitable for the drying needs of heat-sensitive materials. (4) Based on the cloud-edge collaborative architecture and blockchain trusted evidence storage technology, we will build an integrated module for formula production management, energy consumption audit, quality traceability and remote operation and maintenance evidence collection, fill the gap in the traditional system in the whole process traceability and intelligent manufacturing adaptability, realize the whole chain closed loop of "energy flow-quality-traceability", and meet the digital and standardized management needs of modern production lines.
[0019] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description
[0020] Figure 1 This is a schematic diagram of the overall structure and energy flow / air path of a digital twin-driven drying system according to the present invention; Figure 2 A schematic diagram of the layered air supply and return system and sensor / vision arrangement of the drying unit; Figure 3 This is a module architecture diagram for the intelligent control and digital management subsystem. Figure 4 Flowchart for multiphysics modeling and online calibration of the PINN-Transformer coupled digital twin model; Figure 5 A data-driven framework diagram for intelligent regulation of energy flow and quality coordination; Figure 6 A schematic diagram of a state machine for working condition identification and mode switching; Figure 7This is a schematic diagram of the architecture linking cloud-edge collaborative computing, trusted evidence storage, and MES / SCADA / Industrial IoT platform. Figure 8 This is a flowchart of the energy flow-quality collaborative digital twin control method of the present invention. Detailed Implementation
[0021] To make the objectives, technical solutions, and advantages disclosed in the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings and examples. It should be understood that the specific embodiments described herein are only used to explain the embodiments of the present invention and are not intended to limit the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of this application. Examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout.
[0022] It should be noted that the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, such as a process, method, system, product, or server that includes a series of steps or units, not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such process, method, product, or device.
[0023] The following is in conjunction with the appendix Figures 1-8 The embodiments of the present invention will be described in detail below.
[0024] Example System composition and data acquisition: such as Figure 1 and Figure 2 As shown, the system includes a carbon dioxide heat pump power supply subsystem, a heat recovery air circulation and stratified air supply drying subsystem, and a drying device. The carbon dioxide heat pump power supply subsystem supplies heat to the air circulation and drying subsystem. After waste heat recovery and dehumidification, the air enters the drying device in a stratified manner, thereby achieving uniform drying of materials at different levels or in different areas. To meet the needs of digital control, sensors and acquisition units are arranged on the environmental side, the air side, the heat pump side, and the material side, respectively, and industrial cameras are placed at key drying locations to acquire material image information.
[0025] The multi-source heterogeneous process data perception and fusion module performs anomaly removal, missing data completion, unit unification and time alignment on the collected data, forming a unified input dataset that includes heat pump thermodynamic parameters, drying oven temperature and humidity field, air volume distribution, energy consumption and material moisture content / image features, and outputs the feature vector required for operating condition identification.
[0026] The multi-source heterogeneous process data sensing and fusion module has an acquisition frequency of 10~100Hz, and the data integrity after fusion is ≥99.5%, with a time alignment error ≤5ms. Under extremely cold conditions of -20℃, the temperature sensor measurement accuracy is ±0.2℃, the relative humidity sensor measurement accuracy is ±2%RH, and the online moisture content sensor measurement error is ≤1.5%.
[0027] Machine vision and moisture content prediction: such as Figure 2 and Figure 5 As shown, the machine vision and moisture content prediction module preprocesses and detects targets in the image, extracts features such as the apparent color, texture, and size changes of the medicinal materials using Faster-RCNN or SSD, and combines them with a time series model to achieve dynamic moisture content prediction. Cross-calibration with data from online moisture content sensors or weighing units improves prediction accuracy and robustness.
[0028] The industrial camera 116 is placed in a key position of the drying device as the data acquisition terminal for the machine vision and moisture content prediction module.
[0029] The drying uniformity index output by visual prediction is used to characterize the location and extent of high moisture content and low dryness areas, providing a basis for subsequent on-demand energy supply and humidity control strategies.
[0030] PINN-Transformer Coupled Digital Twin and Multiphysics Laws: such as Figure 3 As shown, the PINN-Transformer coupled digital twin model module is an important component of the intelligent control and digital management subsystem; for example... Figure 4 As shown, the dynamic attention mechanism module performs spatiotemporal feature modeling on data such as temperature and humidity, wind speed and material moisture content in multiple zones, and outputs state estimates and short-term trends. The PINN-Transformer coupled digital twin model module embeds heat and mass transfer control equations, air enthalpy-humidity relationship and equipment energy conservation constraints on the basis of data-driven network, and predicts air state parameters, material moisture content distribution, unit energy consumption for moisture removal and quality risk.
[0031] To improve model interpretability and portability, the PINN-Transformer coupled digital twin model module embeds heat and mass transfer and air state constraints into the data-driven network. Typical constraint equations are as follows: (1) Material moisture content diffusion equation: ; (2) Surface convection mass transfer boundary conditions: ; (3) Conservation of air energy in the j-th dry region: ; (4) Enthalpy-humidity relationship of air: , ; In discrete control cycle Δt In this case, digital twins can be assimilated online using a state-space approach: , ; To adapt to alternating temperature and humidity conditions and complex disturbances, the digital twin model adopts an adaptive weighting strategy for physical residuals, adjusting the weights of different physical process constraints online to maintain predictive stability under conditions of condensation, strong dehumidification, or sudden load changes.
[0032] Operating condition identification and mode switching: such as Figure 6 As shown, under alternating temperature and humidity disturbances and severe cold with large temperature rise, the disturbance adaptive operating condition identification and mode switching module triggers state machine switching and updates control target weights when it detects alternating temperature and humidity disturbances, severe cold with large temperature rise, or energy flow offset causing cascade energy mismatch. Through the coordinated optimization of compressor frequency, throttling element opening, fan speed and stratified air supply ratio, the interference of environmental fluctuations on heating stability is suppressed, and the coordinated regulation of energy flow under the coupling of aerodynamic parameters and solid field is realized.
[0033] The disturbance adaptive operating condition identification and mode switching module constructs a disturbance index D(t) and a frost risk index F(t) to trigger operating condition identification and mode switching, wherein: In the formula, Let be the disturbance index at time t; For a moment t Frosting risk indicators; The inlet air temperature; The relative humidity of the inlet air; For the measurement of air supply volume or wind speed; This refers to the surface temperature of the evaporator or heat exchanger. This corresponds to the dew point temperature; For indicator functions; , , , These are the weighting coefficients for each component of the disturbance index; The cumulative weighting coefficient for frost risk; , and These are the normalized reference values for temperature, relative humidity, and air volume, respectively. The sampling interval; For the time window of frosting accumulation; The threshold for the disturbance index; The threshold for frost risk; This is the duration threshold; When satisfied > And the duration exceeds When the conditions are met, the system switches to high-disturbance control mode; when the conditions are met... > When this happens, the system switches to either defrost suppression or defrost scheduling mode.
[0034] Cloud-edge collaboration and trusted traceability: such as Figure 7 As shown, the cloud-edge collaborative computing module completes data fusion, rapid control, and security interlocking at the edge, and completes model training, digital twin inference, and group control scheduling in the cloud; the operation data trusted storage module generates hashes for key operation data, quality data, and control decisions and stores them on the blockchain.
[0035] By connecting with MES / SCADA / Industrial IoT platforms, the system can realize process formula distribution, batch management, energy consumption auditing, anomaly warning and remote operation and maintenance certification, thus forming a digital closed-loop capability that can be used for intelligent manufacturing system integration and operation management.
[0036] Control method flow: such as Figure 8 As shown, the control method is executed according to the above steps; the system completes data acquisition, digital twin prediction, energy flow-quality co-optimization and actuator control command issuance in each control cycle, and completes reliable evidence storage and manufacturing execution linkage at key nodes, thereby achieving stable heating, energy flow matching and high-quality drying throughout the drying cycle.
[0037] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the technical solutions of the present invention, and these modifications or equivalent substitutions cannot cause the modified technical solutions to deviate from the spirit and scope of the technical solutions of the present invention.
Claims
1. A digital twin-driven drying system, characterized in that, include: Carbon dioxide heat pump energy supply subsystem, air circulation and heat recovery subsystem, drying device and intelligent control and digital management subsystem; The carbon dioxide heat pump energy supply subsystem includes a compressor, an air cooler, a throttling element, and an evaporator, which is used to provide heat to the drying unit and form a heat exchange coupling with the air circulation and heat recovery subsystem; The air circulation and heat recovery subsystem includes a return air duct, a fresh air duct, a dehumidification duct, a waste heat recovery unit, a blower, a return air blower, a dehumidification blower, and a stratified air supply actuator, which is used to recover waste heat and condense and dehumidify the return air and supply air to the drying unit in stratified manner. The intelligent control and digital management subsystem is electrically connected to the carbon dioxide heat pump energy supply subsystem and the air circulation and heat recovery subsystem to generate coordinated control commands.
2. The digital twin-driven drying system according to claim 1, characterized in that: The intelligent control and digital management subsystem includes a multi-source heterogeneous process data perception and fusion module, a machine vision and moisture content prediction module, a PINN-Transformer coupled digital twin model module, an energy flow-quality collaborative intelligent regulation module, a disturbance adaptive operating condition identification and mode switching module, a dynamic attention mechanism module, a cloud-edge collaborative computing module, and a reliable operation data storage module. The output of the PINN-Transformer coupled digital twin model module includes air state parameters, spatial distribution of material moisture content, energy consumption per unit of moisture removal, drying rate, and quality risk indicators, which are used to provide feedforward predictions for the energy flow-quality collaborative intelligent control module. The dynamic attention mechanism module is used to jointly model the multi-layer / multi-zone temperature and humidity field, air volume distribution, and material moisture content distribution to form spatiotemporal features for digital twin prediction and control optimization. The cloud-edge collaborative computing module includes an edge controller deployed at the drying site and a cloud server; the edge controller is used to perform data fusion, real-time inference of digital twins, and control output; the cloud server is used to perform model training, recipe optimization, and global parameter tuning, and synchronizes parameters with the edge controller. The trusted evidence storage module for operational data is used to generate hash values for operational data, quality data, and control decisions, and write them into the consortium blockchain or private blockchain ledger in chronological order for verification and traceability. The intelligent control and digital management subsystem interfaces with SCADA systems, MES systems, or industrial IoT platforms via OPC UA, Modbus, or MQTT protocols for process recipe distribution, batch management, and equipment status monitoring.
3. The drying system driven by a digital twin according to claim 2, characterized in that: The multi-source heterogeneous process data sensing and fusion module includes a temperature sensor, a relative humidity sensor, a wind speed / air volume sensor, a pressure sensor, an online moisture content sensor, an energy consumption electrical parameter acquisition unit, and a material weight acquisition unit. It is used to acquire the operating status and material status data of the carbon dioxide heat pump power supply subsystem and the air circulation and heat recovery subsystem.
4. The digital twin-driven drying system according to claim 3, characterized in that: The machine vision and moisture content prediction module includes an industrial camera and an edge inference unit. It uses a target detection and texture feature extraction model to identify the appearance features of materials and outputs online prediction results of material moisture content and drying uniformity.
5. A digital twin-driven drying system according to claim 4, characterized in that: The energy flow-quality collaborative intelligent control module generates a hierarchical energy flow priority allocation scheme based on a reinforcement learning strategy, and combines model predictive control to constrain and optimize compressor frequency, throttling element opening, fan speed, hierarchical air supply valves, and delivery cycle time.
6. A digital twin-driven drying system according to claim 5, characterized in that: The disturbance adaptive operating condition identification and mode switching module constructs disturbance indices based on ambient temperature and humidity fluctuations, material moisture content gradients, and energy flow distribution offsets. It adaptively switches between closed-loop circulation, open-loop dehumidification, deep condensation dehumidification and energy replenishment, and frost suppression / defrosting scheduling modes, and imposes constraints on pressure change rate, exhaust temperature, and heat supply fluctuations.
7. A method for coordinated energy flow and quality control, employing a digital twin-driven drying system as described in any one of claims 1-6, characterized in that, Includes the following steps: S1. Collect operating parameters of the carbon dioxide heat pump power supply subsystem, temperature, humidity and air volume parameters of multiple layers / zones in the drying device, energy consumption parameters and material status data, and perform data cleaning and synchronization. S2. Perform target detection and texture feature extraction on the collected image data to obtain the material appearance features and output the online prediction results of material moisture content and drying uniformity; S3. Use dynamic attention mechanism to perform spatiotemporal feature fusion on the multi-source data obtained in S1 and S2 to form a feature vector for twin reasoning; S4. Based on the constraints of heat and mass transfer and energy conservation, drive the PINN-Transformer coupled digital twin model to predict air state parameters, spatial distribution of material moisture content, energy consumption per unit of moisture removal, and quality risk. S5. Based on the predictions obtained in S4, construct a multi-objective optimization problem of energy consumption, humidity control, and quality. Key control variables were screened through feature importance analysis, and a mapping relationship was established between heat pump COP, material apparent characteristics and macroscopic drying rate, and the weights of multiple objectives were dynamically adjusted. S6. Based on the cooperative strategy of reinforcement learning and model predictive control, hierarchical energy flow priority allocation and multi-actuator cooperative control commands are generated, and operating conditions are identified and modes are switched according to disturbance indicators. S7. Reliably store operational data, quality data, and control decisions, and associate them with batch and formulation information.
8. The energy flow-quality coordinated control method according to claim 7, characterized in that, The specific content of the PINN-Transformer coupled digital twin model driven by heat and mass transfer and energy conservation constraints in S4 is as follows: heat and mass transfer and air state constraints are embedded in the data-driven network. The material moisture content diffusion equation, surface convection mass transfer boundary conditions, zoned air energy conservation relationship and air enthalpy-humidity relationship are used as physical constraint equations in the training and updating process of the digital twin model, so as to achieve high-precision coupled prediction of temperature field, humidity field, moisture content evolution and system energy flow state during the drying process. The diffusion equation for the moisture content of the material is expressed as follows: ； In the formula, Moisture content of the material; The equivalent diffusion coefficient; This is the spatial second derivative of the moisture content, representing the diffusion and transfer of moisture within the material. The surface convection mass transfer boundary condition is expressed as follows: ； In the formula, The moisture content gradient is along the direction of the outer normal to the material surface. The surface convective mass transfer coefficient; Moisture content on the surface of the material; To achieve equilibrium moisture content; No. The energy of air in the dry zone is conserved, as expressed below: ； In the formula, For the first j Air mass flow rate in the dry zone; The specific heat capacity of dry air at constant pressure; and The first j Inlet and outlet air temperatures of the drying zone; For the first j Heat exchange in the drying zone; The latent heat of vaporization of water; For the first j Mass flow rate of water evaporated in the drying zone; The enthalpy-humidity relationship of air is expressed as follows: ； ； In the formula, Moisture content of humid air; It is the partial pressure of water vapor; This is the total pressure of the moist air; Specific enthalpy of moist air; The specific heat capacity of dry air at constant pressure; Air temperature; The latent heat of vaporization of water at a reference temperature of 0℃; The specific heat capacity of water vapor at constant pressure; The prediction model is obtained by jointly optimizing the data fitting loss and the physical residual loss. The loss function is constructed as follows: ； In the formula, This is the total loss function; This is the loss term for data fitting; This is the physical residual loss term; This is the regularization loss term; This is the weighting coefficient for physical residual loss; The regularization loss weight coefficient; Among them, data fitting loss The expression is as follows: ； In the formula, The number of samples; For the first i Predicted values of each output quantity; For the first i Measured values of each output quantity; Among them, physical residual loss The expression is as follows: ； In the formula, Number of physical configuration points; For physically constrained residual operators; For the first j Predicted physical field variables for each configuration point; For the first j The spatial coordinates of each configuration point; For the first j The time coordinates of each configuration point; Among them, the smoothing regularization term The expression is as follows: ； In the formula, For the model k One trainable parameter; This represents the total number of trainable parameters. Set the discrete control period Δ t The digital twin model uses a state-space representation for online assimilation, expressed as follows: ； ； In the formula, for t The system state vector at any given time; for t Control the quantity at all times; for t The amount of disturbance at any given moment; for t The system output vector at each time step; This is the state transition function; For observation mapping function; The PINN-Transformer coupled digital twin model employs an adaptive weighting strategy for physical residuals to adjust the weights of different physical process constraints online, maintaining prediction stability under conditions such as condensation, strong dehumidification, or sudden load changes.
9. The energy flow-quality coordinated control method according to claim 8, characterized in that, The specific details of dynamically adjusting multi-objective weights in S5 are as follows: The energy flow-quality collaborative intelligent control module integrates heat pump thermodynamic parameters, drying oven temperature and humidity field, and material dryness gradient to construct a deep learning-driven energy flow control model; it designs a hierarchical energy flow priority allocation strategy based on a reinforcement learning framework and performs multi-objective optimization. The expression for constructing the partition energy flow priority is as follows: ； In the formula, For the first j Energy flow priority index for the drying zone; For the first j Moisture content deviation or dryness gradient in the drying zone; For the first j Quality risk indicators for the drying area; The heating demand index for the j-th dry zone; Weighting for moisture content deviation; As a quality risk weight; Weighting based on heating demand; j Number the drying zones; The objective function expression for model predictive control optimization is as follows: ； In the formula, The model predicts the control objective function; N To predict the length of the time domain; k For prediction steps; For the first t+k Step-by-step prediction output; For reference trajectory; Q The output error weighting matrix; For the first t+k Step-by-step control increment; R To control the incremental weighting matrix, This is the penalty coefficient for the variance of moisture content; For the first t+k The variance term for the predicted moisture content state parameters; For the first t+k Predicting moisture content state parameters; The constraints that the model predictive control optimization must satisfy are as follows: And meet the safety interlock constraints; In the formula, and These are the lower and upper limits of the control quantity, respectively; and These are the lower and upper limits for controlling the increment, respectively; The expression for constructing the reinforcement learning reward function is as follows: In the formula, Let t be the reward function value; Let t be the system energy consumption or the energy consumption per unit of water removal at time t. Energy consumption penalty weight; Let t be the average moisture content at time t; The target moisture content; This is the variance term for the moisture content of each drying zone; The quality penalty term at time t; Weighted for average moisture content tracking; The weight for the moisture content variance penalty; Weighting for quality penalties.
10. The energy flow-quality coordinated control method according to claim 9, characterized in that, The specific details of operating condition identification and mode switching based on disturbance indicators in S6 are as follows: A disturbance index is constructed based on temperature and humidity fluctuations and material moisture content gradients. When the disturbance index exceeds a preset threshold, the humidity control weight and quality weight are adaptively increased, and energy flow is preferentially allocated to areas with high moisture content to improve drying uniformity and quality consistency. The mode switching includes closed-loop mode, open-loop dehumidification mode, deep condensation dehumidification and energy replenishment mode, and defrost / defrost scheduling mode. During the mode switching process, constraints are imposed on the pressure change rate, exhaust temperature and heat supply fluctuations. The expressions for constructing the disturbance index and the frost risk index are as follows: ； In the formula, Let be the disturbance index at time t; The frost risk index at time t; The inlet air temperature; The relative humidity of the inlet air; For the measurement of air supply volume or wind speed; This refers to the surface temperature of the evaporator or heat exchanger. This corresponds to the dew point temperature; For indicator functions; , , , These are the weighting coefficients for each component of the disturbance index; The cumulative weighting coefficient for frost risk; , and These are the normalized reference values for temperature, relative humidity, and air volume, respectively. The sampling interval; For the time window of frosting accumulation; The threshold for the disturbance index; The threshold for frost risk; This is the duration threshold; When satisfied > And the duration exceeds When the conditions are met, the system switches to high-disturbance control mode; when the conditions are met... > When this happens, the system switches to either defrost suppression or defrost scheduling mode.